bert-eval.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "b3aaa9d1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "There are 2 GPU(s) available.\n",
      "Device name: GeForce RTX 2080 Ti\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd \n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import torch \n",
    "from torch.utils.data import TensorDataset, DataLoader, SequentialSampler\n",
    "from transformers import AutoTokenizer\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score\n",
    "\n",
    "from loop_train_berts import (\n",
    "    BertClassifier,\n",
    "    preprocessing_for_bert,\n",
    "    bert_predict\n",
    ")\n",
    "\n",
    "from utils import plot_roc\n",
    "\n",
    "batch_size=8\n",
    "\n",
    "if torch.cuda.is_available():       \n",
    "    device = torch.device(\"cuda\")\n",
    "    print(f'There are {torch.cuda.device_count()} GPU(s) available.')\n",
    "    print('Device name:', torch.cuda.get_device_name(0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "1472bdc1",
   "metadata": {},
   "outputs": [],
   "source": [
    "positive = pd.read_csv('data/positive.tsv', sep='\\t', index_col=0)\n",
    "positive['target'] = 1\n",
    "negative = pd.read_csv('data/negative.tsv', sep='\\t', index_col=0)\n",
    "negative['target'] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "e044b5f6",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = positive.append(negative)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "72d4ed0d",
   "metadata": {},
   "outputs": [],
   "source": [
    "data['concat'] = data.Title.map(str) + \" \" + data.Abstract.fillna(' ').map(str)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "cb0fe3e2",
   "metadata": {},
   "outputs": [],
   "source": [
    "data['bert'] = data['concat'].apply(lambda x: x.lower())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "ae76659a",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(data.bert, data.target, test_size=0.1, random_state=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e8bc6272",
   "metadata": {},
   "outputs": [],
   "source": [
    "prefix = 'train-'\n",
    "models = ['allenai-scibert_scivocab_uncased.pkl', \n",
    "          'dmis-lab-biobert-v1.1.pkl', \n",
    "          'allenai-biomed_roberta_base.pkl', \n",
    "          'emilyalsentzer-Bio_ClinicalBERT.pkl', \n",
    "          'microsoft-BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext.pkl']\n",
    "tokenizers = ['allenai/scibert_scivocab_uncased', \n",
    "              'dmis-lab/biobert-v1.1', \n",
    "              'allenai/biomed_roberta_base', \n",
    "              'emilyalsentzer/Bio_ClinicalBERT', \n",
    "              'microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext']\n",
    "names = ['scibert', \n",
    "         'biobert', \n",
    "         'biomed-roberta', \n",
    "         'bioclinical-bert', \n",
    "         'pubmed-bert']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "65d7b114",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "scibert 0.9767769176636172\n",
      "biobert 0.9774806474313863\n",
      "biomed-roberta 0.9866291344123856\n",
      "bioclinical-bert 0.9746657283603096\n",
      "pubmed-bert 0.9795918367346939\n"
     ]
    }
   ],
   "source": [
    "probs_df = pd.DataFrame()\n",
    "\n",
    "for name, model, tokenizer in zip(names, models, tokenizers):\n",
    "\n",
    "    tokenizer = AutoTokenizer.from_pretrained(tokenizer, do_lower_case=True)\n",
    "\n",
    "    test_inputs, test_masks = preprocessing_for_bert(tokenizer, X_test)\n",
    "    test_labels = torch.tensor(y_test.values)\n",
    "\n",
    "    batch_size = 8\n",
    "\n",
    "    test_data = TensorDataset(test_inputs, test_masks)\n",
    "    test_sampler = SequentialSampler(test_data)\n",
    "    test_dataloader = DataLoader(test_data, sampler=test_sampler, batch_size=batch_size, num_workers=1)\n",
    "\n",
    "    bert_classifier = torch.load('./models/' + prefix + model)\n",
    "    \n",
    "    probs = bert_predict(bert_classifier, test_dataloader, device)\n",
    "    y_pred = np.where(probs[:, 1] >= 0.5, 1, 0)\n",
    "    \n",
    "    probs_df[name] = probs[:, 1]\n",
    "    \n",
    "    print(name, accuracy_score(y_test, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "630f1359",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmkAAAHwCAYAAAD98PjEAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAABnnUlEQVR4nO3dd3xVRf7/8dcnjTQ6QUG60kMCCCiCFBFQUVFxFcuKoquirO6qLLKudW37U9de1rboqgFFQb+KCCqIokhRSgQRS4BQQ2ghENLm98e9uV5Sbwg3uZD38/GI5Jwzc87nTC7k48ycM+acQ0RERERCS1hNByAiIiIiJSlJExEREQlBStJEREREQpCSNBEREZEQpCRNREREJAQpSRMREREJQUrSREKImf1gZoNqOo5QYWZ/N7OXa+jak83s/pq49uFmZpeZ2exDrHtIn0kzO9/MNpjZXjPrcSjXFqntlKSJlMHM0sxsv/eXzBbvL+34YF7TOdfVOTcvmNcoYmZ1zOwhM1vvvc+1ZjbBzKw6rl9KPIPMLN1/n3PuQefcNUG6npnZTWaWambZZpZuZu+YWbdgXO9Qmdk9ZvZGVc7hnHvTOTcsgGuVSEyr8Jl8FBjvnIt3zn1/CPWLxzbPzHK8fx93m9l8/5+Vt53yvMeLvnb5HXfen/NeM9toZv82s3Az+9ivfJ6Z5fptv1DVuEWqQkmaSPnOcc7FA92BHsCkmg2n8swsooxD7wBDgLOAusAfgWuBJ4MQg5lZqP178yRwM3AT0AjoAMwARhzuC5XzMwi6Grx2a+CHQ6loZuFlHBrv/fvYGJgH/K/Y8anepLDoq0Gx48ne+gOBi4Gxzrkzi8oDbwL/z6/+9YcSv8jhEmr/aIqEJOfcFuATPMkaAGZ2spl9bWa7zGy5/5CQmTUys/+a2SYz22lmM/yOnW1my7z1vjazJL9jaWZ2upk19/ZuNfI71sPMtptZpHd7rJmt9p7/EzNr7VfWmdmNZrYWWFv8fsxsCDAMGOWcS3XO5TvnFgKXAzea2QnecvO8vW2LvL0X7xeLqbw2mGdmD5jZAmAf0M7MrvLGnGVmv5rZdd6yccDHQHO/Xozm/r1IZtbGe19jvL1/283sDr/rxZjZa972WG1mfyveM+dXtj1wI3CJc+5z59wB59w+b4/Tw35FG5rZR954vzWz4/3O8aR5hvP2mNlSMzvV79g9ZjbNzN4wsz3AlWbWx8y+8bbVZjN7xsyi/Op0NbM5ZrbDzLaaZ6j3DODvwMXeNlnuLVvfzF7xnmejmd1flNiY2ZVmtsDMHjezHcA93n1feY+b99g27890hZklmtm1wGXA37zX+j//z6T3+3BvXL9422SpmbUs1rZ1zGwvEA4sN7NfvPs7ez8Tu8wzhHquX53JZva8mc00s2xgcGk/tyLOuXxgCtClvHLl1P8ZWIDf3+fizKyJmX3ojXeHmX1pofc/GnKU0wdOJABm1gI4E/jZu30c8BFwP55emNuAd80swVvlf0As0BVoCjzurdcTeBW4Dk9vwH+AD8ysjv/1nHObgG+AUX67LwWmOefyzOw8PL+8LwASgC+BlGJhnwecROm/yIYC3zrnNhS77rdAOp4etiJXAGOB5kA+8FSAbQC/987VBdYB24CzgXrAVcDjZtbTOZeNp303+fVibColboD+QEdvjHeZWWfv/ruBNkA77/1dXkZ9vHXTnXOLyikDcAlwL9AQz8/+Ab9ji/H8km8EvAW8Y2bRfsdHAtOABnh6aAqAvwJNgL7eGG4AMLO6wKfALDztfALwmXNuFvAgv/cQJXvP/Rqen8UJeHp4hwH+w8InAb/i+ez5x4y37AA8PYcN8PQoZTrnXuTgnqRzSmmPW7xtchaen+FYPAm4jzfhLZoWkOycO948/2Pxf8Bsb0x/Bt40s45+VS/1xloX+KqUa/t4k9vLgIXllSunfifgVLx/n8twK56/CwnAMXj+vmkdRalWStJEyjfDzLKADXgSjLu9+y8HZjrnZjrnCp1zc4AlwFlm1gxPwnG9c26ncy7POfeFt96fgP845751zhU4514DDgAnl3Ltt/D8QsTMDBjt3QeeJO8h59xqb6/Cg0B38+tN8x7f4ZzbX8q5mwCby7jnzd7jRf7n7W3LBu4ELvL22pTZBn51JzvnfvD21OU55z5yzv3iPL7A80v7VCrnXufcfufccmA5UJS4XAQ86G3zdLzJZBkal3P//t5zzi3ytvGb+PW8OOfecM5leu/tMaAOnuSxyDfOuRnettnvnFvqnFvoLZ+GJ0Ef6C17NrDFOfeYcy7HOZflTZhLMLNj8Hy+/uKcy3bObcPzPwGj/Yptcs497b1W8Z9/Hp5EqBNg3s9QIG0BnkTwH865Nd6f4XLnXGYA9U4G4oGHnXO5zrnPgQ/xfr693nfOLfC2V04Z53nKPPPM9gLj8STQ/i7y9nwVfc0tdvw7b0/dajzDpc+VE3Me0Axo7f3sfumcFruW6qUkTaR85znn6gKD8PxSK0peWgN/8P+FgKeHpxnQEtjhnNtZyvlaA7cWq9cST+9JcdOAvmbWHE/Ph8PTY1Z0nif9zrEDMOA4v/oH9ZIVs90ba2maeY+Xdp51QCSediivDUqNwczONLOF3uGjXXgSOv+EMBBb/L7fh+eXP3ja0P965d1/JmXffyDXwsxuNc+w6m7vvdTn4Hspfu8dvMNnW7xDoA/6lW8J/BJAPOBp90hgs1+7/wdPD1Wp1/bnTZCeAZ4FtprZi2ZWL8BrVyZOf82BDc65Qr996wj881rkJu88s2g8ie0085suALztnGvg91V82LQnnp/hxXh6G+PKudYjeHraZptnaP72AOITOayUpIkEwNvrMxnPE2vg+YXyv2K/EOK885k2AI3MrEEpp9oAPFCsXqxzrvhQJc65XXh6mi7CMxSU4vd/8huA64qdJ8Y597X/Kcq5pU+Bk0qZT9QHzy/iz/12+5dphaeHYXsFbVAiBu+Q7rt42vAY7y/bmXiSy4riDcRmoEUZcRf3GdDCzHodyoXMM/9sIp6fTUPvvezm93uBkvfzPPAj0N45Vw/P8FlR+Q3A8ZSu+Hk24Ol9beLX7vWcc13LqXPwCZ17yjl3Ip7h+A7AhEDqVRBneTYBLYvN6WoFbPQPK9CTeXvbvsSTRFX41Gqxus459zae6QR3lVMuyzl3q3OuHXAOcIt55nKKVBslaSKBewIYambdgTeAc8xsuHcydbR5XiHRwjt09DHwnJk1NLNIMxvgPcdLwPVmdpJ3AnecmY3wzkkqzVt45oSN4vehToAXgElm1hV8E8n/EOiNOOc+xZOovGueCevhZnYyniG9551z/g8bXG5mXcwsFrgPz7y4gvLaoIzLRuEZEswA8s3sTA7+BbsVaGxm9QO9j2LextMmDb3z5caXVdB7f88BKd6Yo7zxjw6wx6QunjlhGUCEmd2FZ45WRXX2AHu9c6LG+R37EDjWzP5inon3dc3sJO+xrUCbogTH+/maDTxmZvXMLMzMjjezgQTAzHp7P3+RQDaQg2e+XNG12pVT/WXgn2bW3vv5TTKzxgFc9lvvtf7m/fswCE/iMyWQmMu4j7545lse0hOkwMPAtWZ2bBnnP9vMTvBONdiDp40KSisrEixK0kQC5JzLAF4H7nSeCfcj8fSGZODpYZjA73+n/oinx+lHPHPZ/uI9xxI889KeAXbi6Qm4spzLfgC0B7Z652AVxTId+BcwxTt0lopnnlJljALm4pmsvhdP0vUKnknd/v6HpxdxC55hppu8MVTUBgdxzmV5676N594v9d5f0fEf8Tz88Kt3GK+0IeDy3IdnovdveHoKp+HpcSrLTfw+7LcLzzDe+XgmuFfkEzyJ+E94hu1yqHi47jY895yFJ1mfWnTA2zZD8SQuW/A8kVs0VPeO989MM/vO+/0VeJLeVXjachqBDd+CJ5l8yVtvHZ6h36Ie4leALt72n1FK3X/j+fnNxpO4vALEVHRB51wucC6ez+h2PAnyFd6feWU8Y96nf/F8Lv/hnPvY73jRU7D+X01LO5FzbiXwBb/3IhbXHs/naC+eXrfnXDW9w1CkiGkepIiUxczmAW8452rkrf9VYWbjgNHOuYB6mEREQo160kTkqGBmzcysn3f4ryOeVyhMr+m4REQOVY29BVtE5DCLwvOUY1s8w5dTKP8VCyIiIU3DnSIiIiIhSMOdIiIiIiFISZqIiIhICDqq5qQ1adLEtWnTpqbDEBEREanQ0qVLtzvnEso6flQlaW3atGHJkiU1HYaIiIhIhcxsXXnHNdwpIiIiEoKClqSZ2atmts3MUss4bmb2lJn9bGYrzKyn37EzzGyN95gWtRUREZFaJ5g9aZOBM8o5fiaeZTfaA9fiWXwYMwvHs0zLmXjWZbvEzLoEMU4RERGRkBO0JM05Nx/YUU6RkcDrzmMh0MDMmgF9gJ+dc79613ub4i0rIiIiUmvU5IMDx3HwgsTp3n2l7T+pGuMSEREJuszNO3nw0ifYnbGnpkORw6QQR25BLnB4FgqoySTNStnnytlf+knMrsUzXEqrVq0OT2QiIiJBtnxuKiu+WFXTYUgIq8kkLR1o6bfdAtiEZ/290vaXyjn3IvAiQK9evbTGlcgR7MdftvDruu01HUa1cDi25WyjgMJSj6+a+h1bvt9YzVFJdcrZtR+AZr1aknxVyQGjzOwD7MzJqe6wpAryCvM4UHiAY+scg1nFM8o+nTSt3OM1maR9AIw3syl4hjN3O+c2m1kG0N7M2gIbgdHApTUYp4hUk1/XbSdzVzaNG8TVdChBt69gH+n7S///T1fgWPnGksM1YiIhLqJFHfY1zS2xf0tmDvtjComJ1NuyjhyRRFs0cTGNsFIHBisnaEmamaUAg4AmZpYO3A1EAjjnXgBmAmcBPwP7gKu8x/LNbDzwCRAOvOqc+yFYcUrt9M3/LeHTN+bjXOV/C2btzSF7X8l/UI9WeYW5FB5CO5UlJz+fAwUFpR4ryHeERxhx9esctuuFqkIKOVCQQ3xEPJFhkQcdc875Jn+c88qYmglQqkVYZDhNux1HWETJRKzJvmwSYuO4sEtiDUR2MFewCQoyajqMo849fyz/eNCSNOfcJRUcd8CNZRybiSeJEwmK//4jhd9Wrq/pMKQMu2s6gGq0s5xjsU3iad6nTXWFIiEmITaOjk3KXDGoeuV8givMqukoap2jalkoCZ6amCvknOP76d+yKz3zsJ1zb24e+/Jy2fbLFgDaXnoSdRpVbmjtQE4+daIjaNG20WGLK9gOFOSS7/IqXa/AFbI9N5MmUY2Ij4g/LLHsOXCAenXq0Oe44w7L+Y5kZka4lf3PcIde7WjW9phqjKhmOOegYAO4AzUdSgjag8sLgac/XS4W2RGiTq3pSI4yN5d7VEmaBKQm5grt2riDL577JHgXMEg4uR1RDWIrXbXJMXU55rj6QQjq8HPAsp3LKCSywrKlaURz2sS3pWFkg8MWU8cmCXRrevQnHxIgtxu3f0ZNRyEVMIvHwir/76UcOiVpIcY5x5zXv2DLb9tqOhQAtu/cy85d+9iXk0tsdBTt2zYN+jW3Zu8lc98+9m3fC0Bs07r0+FP/w3Luvbm5xEdF0ePY5rTs2JyepycdlvMGyhVmQ2F5A1xl21eQQ3b+/kOqGx7vaBt3HO1im1e6bpiFERceg1nVJ8H+Lg+Xn34YzydHNOfpKbI6p0K4XqUUkszAGtR0FLWOkrQQs25VOo9c9WxNh1GmhTVwzXrHNSDxkj6H7Xw12ouT8wmu4NCSkwU7MthbxoT7QMTlp1M3/xB7QvP0oKFUg7DGWHjjmo5CJGQoSQsx+7I8PSVNjmvEmVcPqeFoYK23R686etCKrMrwXLNLQlPMjP4X9KFtl9YB1fU8FbcTKnga0RUc+jy3AldwyD1a5O0CawiRvStfNXwZx8bU44S4lhUXLsYwGkbVxbM0rkgIsggI0xC4iD8laSGqSYvGXHHPRdV6zdIeDrCk1jRuEMdZp1XfI+DTVqUCHNpj53nf4w58dZgjOtiyrN2szznEJA0grBEWseYQKsYQF9WSpnHdDv3aIiJyxFCSJj6lPRzQuEEc7Vo3qdJ5C10heYWBP1mYX5iPI4+cvEOYu5W/E1whRA2tfN0AHQj/idg6+0ms2+bQThCeAFb5ybdmRpMoDQWJiNQWStJqgUBfn1GUoB2OXrOV27ayZrvnxYdr9/7MnrzA36+TtT+LupEZ1NlwqI/jh2FRPx9i3cDO3zDqOFrUOyWI1xARkdpOSdphVFBQwJpFP5ObU/n3URVZt+rwPfFW9Db9X9ZlsGNXNo0qeH1GowaxtGvV5JDewl+8zo8Z28jwvi07tyCXmPCYwHuB6myjTfweOjbqzSF9RK0uFn5s5etVQqOohkE9v4iIiJK0w2jKQzOYfNeUw3KusPCK12orr4ds64FtpO/zLM6cvSefuHoRxHeo5zu+YWcOm/eUXNroq63A1koG63KhYCP+z/9lHTDq1nEc2/gA3WLguDrR9K7XoBInjcXiT8MsupLBiIiIHB2UpB0G+7L2s3VdBr+sSAPguPbNaHLcob+N3sKM88afWWG58l4we6DgAGFmHBN9DMRAixZ1ae2XJG3YuoFod4CGMYdhjUS3B/IyILwJEAVAsxho0yCKE7xrMDar0wCLrMQ8rLB4JWgiIlKrKUmrorzcPK7s8Gd2bv19tcE/3HoOI64N3sR1f2XNIVu2s4BNOcZZzUqP44eYfJrHBP4EpTuwEPJWlnG0AOdisdizgz7MKCIiUlsoSaui7N372Ll1N2FhRqvOLajbKJ5ew7vXdFiHX+EWsDCIOL7Uw0YUhFXtKVARERH5nZK0w6Ruo3heWvnvoJ2/tPlnlV1L0+Wthtylnu8PeN7z5bKXBVg5C8KaYHUGBXw9EREROXRK0o4Qh+UdZgXpnmQrog2E7fLsC28QYOXGEN4u8GuJiIhIlShJO4JU9A6zwsJClmb8l/35WazKzGPjbsjeuuX3Am4fEIbVacn2vEYkxMZh0dW3koCIiIgETknaUSTP7WVD9i/ER8STmRVDeEEEYL8XsDgI8zx1mhAbR8cmCTUTqIiIiFRISdohWjpnOY9d8zw5e3OqdJ7KrgZQ9CZ/B2zav5EDBbmeHrLC7RRSyO7cOrSMa0ejOm1JaBh3aOtfioiISI1TknaIvv3oOzI2ZPq2O5/c4ZDOU967zvwVzT9L3Z5Bxr5sGsfEsCVnG5FhEUQU7oDCnWBRxETEEhvZhPhI9ZSJiIgcyZSkVdGV943m7OuHUq9x3UM+RyDrZf689xf25m1l/S+bADiheTMi69eja/0utI/aAnk/YPHXH3IMIiIiElqUpFVRTHw09ZvUq7hgFTjnWLkrlfCwcHZ5FyrfkmNEh0dTP7IesKX8E4iIiMgRR0laJf33Hyms/HI1m34JfmK0v2A/adnrcN41MdvHn0Bu/XwAzmr2e8+bO/BT0GMRERGR6qUkrRJy9h3grQffO2hf08q8p6yS1u/bwI971gBgZtSNiAd2Be16IiIiEjqUpFWCKywEICo6kodm/YP4BnG07daq0ufxf6KztIcGtuZsY2/+XpZuXs/KjD30bNgdw1i4excZ+7JJiPWUd/m/QeEeKMio4p2JiIhIqFGSdgjCwsNIGtDlkOv7P9FZ2qoBi3YsJr8wn5UZe8jJNfzfdVb0fjPn8iHnQ5zzDIWa9/1nIiIicnRQkhaA7Zt28O6/P2TvruwqnaeoB60oQfN/onPZlo0s3vobOMdv2TtJqNOE4yKPp2nDeP7g964z5w5A/nrI34FzDovqA1HJQFSVYhMREZHQoiQtAB+/9BnT/v1/vu26jeIP6Tz+CVrx3rOvN//Eqh3rqBsdDkBkWCTHxtYr+a6zvOW4Awt/3w6ri1nMIcUjIiIioUtJWgByc3IB6HtuL3oP707iqZ0P+VxlvRPNOUfd6HD+1vtsIiyC+Ih4zKzkCVw+ZmEQeylgYA0OORYREREJXUrSKqHzSR04Z9zwoF4jPiKeyLDICstpDpqIiMjRLaymAxARERGRktSTFmQVvW6jaMH0nfurtlC7iIiIHF2UpAWZ/8MCDRvE0KJlPXIKfk/IUrdtYvu+fdSLjqRelJ7QFBEREQ8laYeRf69ZEf/Xbczd9gW/5P7GvFU5bN7jeRghK6eAutHh9GleDywao5SHBURERKTWUZJ2GPn3mhXxf91GTkEOjes0Ykt+AXU5QMPYOhAL7RrWp0ODhsRExBARph+JiIiIKEk77Mp6xUaRuhF1SagTQUIduLBL2eVERESkdlOSVo792Tls+XUru7btrulQREREpJZRklYG5xzXd7+NTb9s9e3zf7dsefPPqnzt/dPLWTQ9HzRvTURE5KinJK0MhYWFvgStTWJLYuvF0vfcXr7jFc0/q5KCzRDWCMKPLf14WOOqX0NERERCmpK0CoSFGS+t+Hepxyqaf1Yl4S2xOv2Cc24REREJeVpxQERERCQEqSetGq3fuZ9V+zYRS2MSYqs+d01ERESOXkrSqtHG3Tm4fGjdOI6OTRJqOhwREREJYUrSgqxobU6ArJx8WtSN0vvRREREpEKakxZka7ZnkLEvG4C60RG0bhBfwxGJiIjIkUA9aZVU9H60QN+JlluYRw47Ob5ZHaIaxNMypl41RCkiIiJHOiVpleSfoAXyTrSsvD1sO5DB5hyItEgaRjWshihFRETkSKck7RAcyvvRBiUMIDYiNkgRiYiIyNFGc9JEREREQpCSNBEREZEQpCRNREREJARpTlqQFL0fbcf+nIDKO5cL+euAQu+XiIiI1GZK0oKk6P1ojWKiiYuIqrhC3mrcgS98m2bRQYxOREREQp2StCBKiI2jd+tj+W7nrgBKFwBgsReDRYE1CGZoIiIiEuKUpIWasIaYBdDzJiIiIkc1JWnF7M/O4YupX5O9e19NhyIiIiK1mJK0Yma98jnP/eW/vu2omCjfUlBAwMtBiYiIiFSFkrRi9u7yLIbesffxdD65Az1O63bQUlCBLgclIiIiUhVK0srQa3h3rrxvNAAzP089pKWgRERERA6VkrQgcd7/Fnq/K7Occ97S5ZcTERGR2kVJWpD8svcXduftIaJePQDMrPSC+9/BFWypxshERETkSKAkzStn3wE2rt3Mzi27Ds/5CnKICY+hc71O1AmvQ3RYGS+nLdyJhR8LEW3B6ur1GyIiIgIoSfP580mTSPthg2+7zJ6vSogNj6ZTvY4VFww/BovqXeXriYiIyNFDSZpX+k+bAGiT2JK4+rGcOurkGo5IREREajMlacU8t+RfREZF1nQYIiIiUsuFBfPkZnaGma0xs5/N7PZSjjc0s+lmtsLMFplZot+xNDNbaWbLzGxJMOMUERERCTVB60kzs3DgWWAokA4sNrMPnHOr/Ir9HVjmnDvfzDp5yw/xOz7YObc9WDGKiIiIhKpg9qT1AX52zv3qnMsFpgAji5XpAnwG4Jz7EWhjZscEMSYRERGRI0Iwk7TjgA1+2+neff6WAxcAmFkfoDXQwnvMAbPNbKmZXRvEOA+b/QX7WbxjCQszvyXP5dd0OCIiInIEC+aDA6W9w6L4a/UfBp40s2XASuB7oCi76eec22RmTYE5Zvajc25+iYt4ErhrAVq1anW4Yj8k2w9kkr5vI/ER8dQJq0PdyHo1Go+IiIgcuYKZpKUDLf22WwCb/As45/YAVwGY58Vkv3m/cM5t8v65zcym4xk+LZGkOedeBF4E6NWrV42srZS+L52MA9tJ3baVVdv30LXecSRExNM4Kq4mwhEREZGjQDCHOxcD7c2srXleoz8a+MC/gJk1sN9fsX8NMN85t8fM4sysrrdMHDAMSA1irFWyJusn1u/bwNodO8nLiyAyLJKE2Dg6Nkmo6dBERETkCBW0njTnXL6ZjQc+AcKBV51zP5jZ9d7jLwCdgdfNrABYBVztrX4MMN371v8I4C3n3KxgxVqWH3/Zwq/rtpO5K5vGDcruFUvbsY/MLGgW2ZaE+nFc2CWxzLIArjAb8lKBAu+XiIiIyMGC+jJb59xMYGaxfS/4ff8N0L6Uer8CycGMLRD+CVq71k3KLLdx935ycsNo1yTA3rOCX3C532IWBhiENT58QYuIiMhRQSsOVKBxgzjOOq38njGABtFRFfeg5adD4XYo2OjZETsWC4s9HGGKiIjIUUZJWnU6MAdXmAWAWTSYlp8SERGR0tXqJG1Xxm6m/ut9snfvoyC/sFJ1V27byprtGQDszskjOjY6gFqFWGQnqDMQCMesVje/iIiIlKNWZwlzUxYw7d//59uOqx9LWHhgD7yu2Z5Bxr5sEmLjqB8dSctyHiw4WARmdQ4hWhEREalNanWSlpuTB8CJw5IZMOpkOvQ+nvDw8IDrJ8R6nuT8dOs26kbUDVaYIiIiUgvV6iStyPFJrTnrT6fXdBgiIiIiPkrSgsS5HPA+JPD7zsrNexMREZHaS0lasOyfgSvYVmK3HhYQERGRQChjqKRCV8iBwgPkFnrms+0v2I9zvy8Z6tx+cAXg9mPhx0FU94NPEH5cNUYrIiIiRyolaZU0Ze3nrMzYTFZOAXWjw4nZvB6ABlH1cfkbcPun+8paZEss4viaClVERESOYErSKumXHbvJy4uga6OWtGtUnw4NGwLQtE4CuHQArM7JYLEQ3romQxUREZEjmJK0Q9AkNpZxPQaV2O/yvN9EnICFNarOkEREROQoE9ibW0VERESkWqknrRQ//rKFX9dtJ3NXNo0DXklARERE5PBRT1op/BO0dq2b1HQ4IiIiUgupJ60MjRvEcdZpiTUdhoiIiNRS6kkTERERCUFK0kRERERCkJI0ERERkRCkOWleRU90AnqqU0RERGqcetK8ip7oBPRUp4iIiNQ49aT5OdQnOl3eGihIh8LdQYhKREREaiMlaYdD3hJPgmbRWHgTsPiajkhERESOcLUySXvn0Q/48r2FbN+44/Cc0DkIb4PFnHV4ziciIiK1Xq1M0t568D32euefARzbtmkNRiMiIiJSUq1M0goLCwF4+JN/0PCYBrTt1oqP5/5Qw1GJiIiI/K5WJmlFOvU5gbj6etWGiIiIhJ5anaQFauW2razZngHAnpw8GsdG4FwBFPwGLh84ULMBioiIyFFHSVoA1mzPIGNfNgmxcdSLjqRVgzgoSMftn+krY+ExNRihiIiIHG2UpAUoITaOC7skMndbJtFhdYB8ACzmbAhrBFavZgMUERGRo4qStKqyulhYg5qOQkRERI4yWhZKREREJAQpSRMREREJQUrSKssV4lwOuJyajkRERESOYrV+TtqPv2zh13XbydyVTeMGAbwzLT8V53bgchp6ti08uAGKiIhIrVTrkzT/BK1d6yYVV3AHsLC6WPQQsDpgDYMfpIiIiNQ6tT5JA2jcII6zTksssb/oJbZF70jzsWgssms1RigiIiK1jeaklcM/QevYJKGmwxEREZFaRD1ppSjeg3Zhl5K9bCIiIiLBpJ60UqgHTURERGqaetLKoB40ERERqUnqSRMREREJQUrSREREREKQkjQRERGREKQkTURERCQEKUkTERERCUF6ujMAzh2A3G/A5XsXVo+t6ZBERETkKKckLRAFW3C5K7CwWCAcwrRep4iIiASXkrTKiB4BdX7CwurUdCQiIiJylNOcNBEREZEQpCRNREREJAQpSRMREREJQUrSREREREKQkjQRERGREKQkTURERCQEKUkTERERCUF6T1o5nMuFwp1QuKumQxEREZFaRklaeXI+weX/5rcjssZCERERkdpFSVp53AEsvAlE9QWLgrDGNR2RiIiI1BK1Lkn7ZPJc9mflEBZmhEWEV1zBorGItsEPTERERMRPrUnSCgsLeWXSW7z9yPsAXDLpAmLiokst6ygEnPdLREREpPrVmiRt4f8t5e1H3ic8IpzxT1/N2dcNLbWcK9wFB+aBK8QVhGERrao1ThERERGoRUnatvXbAThj7GllJmgAuGxwhRB+HFanHYS3rKYIRURERH4X1PekmdkZZrbGzH42s9tLOd7QzKab2QozW2RmiYHWPVQRkQHMQwMIPwaLOhELb3q4Li0iIiISsKAlaWYWDjwLnAl0AS4xsy7Fiv0dWOacSwKuAJ6sRF0RERGRo1Ywe9L6AD875351zuUCU4CRxcp0AT4DcM79CLQxs2MCrCsiIiJy1ApmknYcsMFvO927z99y4AIAM+sDtAZaBFhXRERE5KgVzCTNStlX/J0WDwMNzWwZ8GfgeyA/wLqei5hda2ZLzGxJRkZGFcIVERERCR3BfLozHfB/NLIFsMm/gHNuD3AVgJkZ8Jv3K7aiun7neBF4EaBXr156sZmIiIgcFYLZk7YYaG9mbc0sChgNfOBfwMwaeI8BXAPM9yZuFdYVEREROZoFrSfNOZdvZuOBT4Bw4FXn3A9mdr33+AtAZ+B1MysAVgFXl1c3WLGKiIiIhJqgvszWOTcTmFls3wt+338DtA+0roiIiEhtEdSX2YqIiIjIoVGSJiIiIhKClKSJiIiIhCAlaSIiIiIhSEmaiIiISAhSkiYiIiISgpSkiYiIiIQgJWkiIiIiIUhJmoiIiEgIUpImIiIiEoKCuizUkWTltq2s2Z6BK9xJxn5HQlTFdURERESCRT1pXmu2Z5CxLxuAhBijY6P6NRyRiIiI1GbqSfOTEBvHqI4tcPu+x2Ia1XQ4IiIiUoupJ01EREQkBClJExEREQlBStJEREREQpCSNBEREZEQFHCSZmZxwQxERERERH5XYZJmZqeY2SpgtXc72cyeC3pkIiIiIrVYID1pjwPDgUwA59xyYEAwgxIRERGp7QIa7nTObSi2qyAIsYiIiIiIVyAvs91gZqcAzsyigJvwDn2KiIiISHAE0pN2PXAjcByQDnQHbghiTCIiIiK1XiA9aR2dc5f57zCzfsCC4IQkIiIiIoH0pD0d4D4REREROUzK7Ekzs77AKUCCmd3id6geEB7swERERERqs/KGO6OAeG+Zun779wAXBjMoERERkdquzCTNOfcF8IWZTXbOravGmERERERqvUAeHNhnZo8AXYHoop3OudOCFpWIiIhILRfIgwNvAj8CbYF7gTRgcRBjEhEREan1AknSGjvnXgHynHNfOOfGAicHOS4RERGRWi2Q4c4875+bzWwEsAloEbyQRERERCSQJO1+M6sP3Irn/Wj1gL8EMygRERGR2q7CJM0596H3293AYPCtOCAiIiIiQVLey2zDgYvwrNk5yzmXamZnA38HYoAe1ROiiIiISO1TXk/aK0BLYBHwlJmtA/oCtzvnZlRDbCIiIiK1VnlJWi8gyTlXaGbRwHbgBOfcluoJTURERKT2Ku8VHLnOuUIA51wO8JMSNBEREZHqUV5PWiczW+H93oDjvdsGOOdcUtCjExEREamlykvSOldbFCIiIiJykPIWWNei6iIiIiI1JJBloURERESkmilJExEREQlBASVpZhZjZh2DHYyIiIiIeFSYpJnZOcAyYJZ3u7uZfRDkuERERERqtUB60u4B+gC7AJxzy4A2wQqouqVn7WHxpnQy9mXXdCgiIiIiPoEkafnOud1Bj6SGbM7aQ1ZuLgmxcXRsklDT4YiIiIgA5b8nrUiqmV0KhJtZe+Am4OvghlW96kZFcWGXRABcwcYajkZEREQksJ60PwNdgQPAW8Bu4C9BjElERESk1gukJ62jc+4O4I5gByMiIiIiHoH0pP3bzH40s3+aWdegRyQiIiIiFSdpzrnBwCAgA3jRzFaa2T+CHZiIiIhIbRbQy2ydc1ucc08B1+N5Z9pdwQxKREREpLYL5GW2nc3sHjNLBZ7B82Rni6BHJiIiIlKLBfLgwH+BFGCYc25TkOMREREREQJI0pxzJ1dHICIiIiLyuzKTNDN72zl3kZmtBJz/IcA555KCHp2IiIhILVVeT9rN3j/Pro5AREREROR3ZT444Jzb7P32BufcOv8v4IbqCU9ERESkdgrkFRxDS9l35uEORERERER+V96ctHF4eszamdkKv0N1gQXBDkxERESkNitvTtpbwMfAQ8DtfvuznHM7ghqViIiISC1XXpLmnHNpZnZj8QNm1kiJmoiIiEjwVNSTdjawFM8rOMzvmAPaBTEuERERkVqtzCTNOXe298+2h3pyMzsDeBIIB152zj1c7Hh94A2glTeWR51z//UeSwOygAIg3znX61DjEBERETnSBLJ2Zz8zi/N+f7mZ/dvMWgVQLxx4Fs+ToF2AS8ysS7FiNwKrnHPJwCDgMTOL8js+2DnXXQmaiIiI1DaBvILjeWCfmSUDfwPWAf8LoF4f4Gfn3K/OuVxgCjCyWBkH1DUzA+KBHUB+oMGLiIiIHK0CSdLynXMOT4L1pHPuSTyv4ajIccAGv+107z5/zwCdgU3ASuBm51yh95gDZpvZUjO7NoDriYiIiBw1KlxgHcgys0nAH4FTvcOYkQHUs1L2uWLbw4FlwGnA8cAcM/vSObcH6Oec22RmTb37f3TOzS9xEU8Cdy1Aq1YVjsKKiIiIHBEC6Um7GDgAjHXObcHTG/ZIAPXSgZZ+2y3w9Jj5uwp4z3n8DPwGdAJwzm3y/rkNmI5n+LQE59yLzrlezrleCQkJAYQlIiIiEvoqTNK8idmbQH0zOxvIcc69HsC5FwPtzayt92GA0cAHxcqsB4YAmNkxQEfgVzOLM7O63v1xwDAgNcB7EhERETniBfJ050XAIuAPwEXAt2Z2YUX1nHP5wHjgE2A18LZz7gczu97MrvcW+ydwipmtBD4DJjrntgPHAF+Z2XLvtT9yzs2q/O2JiIiIHJkCmZN2B9DbO+yImSUAnwLTKqronJsJzCy27wW/7zfh6SUrXu9XIDmA2ERERESOSoHMSQsrStC8MgOsJyIiIiKHKJCetFlm9gmQ4t2+mGK9YyIiIiJyeFWYpDnnJpjZBUB/PK/VeNE5Nz3okYmIiIjUYmUmaWbWHngUz/vLVgK3Oec2VldgIiIiIrVZeXPLXgU+BEYBS4GnqyUiERERESl3uLOuc+4l7/drzOy76ghIRERERMpP0qLNrAe/L+8U47/tnFPSJiIiIhIk5SVpm4F/+21v8dt2eNbbFBEREZEgKDNJc84Nrs5AREREROR3eimtiIiISAhSkiYiIiISgpSkiYiIiISgCpM087jczO7ybrcysz7BD01ERESk9gqkJ+05oC9wiXc7C3g2aBGJiIiISEALrJ/knOtpZt8DOOd2mllUkOMSERERqdUC6UnLM7NwPO9Gw8wSgMKgRiUiIiJSywWSpD0FTAeamtkDwFfAg0GNSkRERKSWq3C40zn3ppktBYbgWRLqPOfc6qBHJiIiIlKLVZikmVkrYB/wf/77nHPrgxmYiIiISG0WyIMDH+GZj2ZANNAWWAN0DWJcIiIiIrVaIMOd3fy3zawncF3QIhIRERGRyq844Jz7DugdhFhERERExCuQOWm3+G2GAT2BjKBFJCIiIiIBzUmr6/d9Pp45au8GJxwRERERgQqSNO9LbOOdcxOqKR4RERERoZw5aWYW4ZwrwDO8KSIiIiLVqLyetEV4ErRlZvYB8A6QXXTQOfdekGMTERERqbUCmZPWCMgETuP396U5QEmaiIiISJCUl6Q19T7ZmcrvyVkRF9SoRERERGq58pK0cCCeg5OzIkrSRERERIKovCRts3PuvmqLpBqt3LaVNds9r3rLys2lblRUDUckIiIicrDyVhworQftqLBmewYZ+zzPQNSNiqJZ3Xo1HJGIiIjIwcrrSRtSbVHUgITYOC7skkjslpqORERERKSkMnvSnHM7qjMQEREREfldpRdYFxEREZHgU5ImIiIiEoKUpImIiIiEICVpIiIiIiFISZqIiIhICFKSJiIiIhKClKSJiIiIhCAlaSIiIiIhSEmaiIiISAhSklYJ+wv2s79gP9hRu6ypiIiIhIjy1u4UPztzd7IwcxEFroAT4tvVdDgiIiJylFOSFoDs/Gy+3L6AOmF1GNCkP/Uj69d0SCIiInKU03BnAPbm76WgsICeDXsoQRMREZFqoSStEsLUXCIiIlJNlHWIiIiIhCAlaSIiIiIhSEmaiIiISAhSkiYiIiISgpSkiYiIiIQgJWkiIiIiIUhJmoiIiEgIUpImIiIiEoKUpImIiIiEICVpIiIiIiFISZqIiIhICFKSJiIiIhKClKSJiIiIhCAlaSIiIiIhSEmaiIiISAgKapJmZmeY2Roz+9nMbi/leH0z+z8zW25mP5jZVYHWFRERETmaBS1JM7Nw4FngTKALcImZdSlW7EZglXMuGRgEPGZmUQHWFRERETlqBbMnrQ/ws3PuV+dcLjAFGFmsjAPqmpkB8cAOID/AuiIiIiJHrWAmaccBG/y20737/D0DdAY2ASuBm51zhQHWFRERETlqBTNJs1L2uWLbw4FlQHOgO/CMmdULsK7nImbXmtkSM1uSkZFx6NGKiIiIhJBgJmnpQEu/7RZ4esz8XQW85zx+Bn4DOgVYFwDn3IvOuV7OuV4JCQmHLXgRERGRmhTMJG0x0N7M2ppZFDAa+KBYmfXAEAAzOwboCPwaYF0RERGRo1ZEsE7snMs3s/HAJ0A48Kpz7gczu957/AXgn8BkM1uJZ4hzonNuO0BpdYMVq4iIiEioCVqSBuCcmwnMLLbvBb/vNwHDAq0rIiIiUltoxQERERGREKQkTURERCQEKUkTERERCUFK0kRERERCkJI0ERERkRCkJE1EREQkBClJExEREQlBStJEREREQpCSNBEREZEQpCRNREREJAQpSRMREREJQUrSREREREKQkjQRERGREKQkTURERCQEKUkTERERCUFK0kRERERCkJI0ERERkRCkJE1EREQkBClJExEREQlBStJEREREQpCSNBEREZEQpCRNREREJAQpSRMREREJQUrSREREREKQkjQRERGREKQkTURERCQEKUkTERERCUFK0kRERERCkJI0ERERkRCkJE1EREQkBClJExEREQlBStJEREREQpCSNBEREZEQpCRNREREJAQpSRMREREJQUrSREREREKQkrQAZOfvAyAyLLKGIxEREZHaQklaANbtW0f9qPrUjYiv6VBERESkllCSVlx+mudPqwPAztyd7MrdTZvY1phZzcUlIiIitYqSND8u/2dc7lIsMhELbwpAWvY6wsPCaRHbooajExERkdpESZqXK8iEnDlY+LFQZwAAeYV5pO/fyHExzYnSfDQRERGpRkrSAEc+5HwEFgXRZ2EWAUD6/o3kF+bTJrZNzQYoIiIitU6tT9IcDvJSwe2B6DOxMM/DAYWukF/2/kq9yLo0impYw1GKiIhIbVPrkzTyf4OCDKgzAAtv7tv9a/ZvZOVl0bleZz0wICIiItWuVidprnAX5P8C4c0hoptvf05BDqv3/EjT6KY0iz625gIUERGRWqtWJ2m4HM+f4ccc1FuWunsVha6Q5Prd1IsmIiIiNaJ2J2mlyMrLYsO+DZwQfzzxkXp5rYiIiNQMJWnFHCg8AEDT6IQajkRERERqMyVpIiIiIiFISZqIiIhICIqo6QBERKR2ysvLIz09nZycnJoORSSooqOjadGiBZGRlVu9SEmaiIjUiPT0dOrWrUubNm30JL0ctZxzZGZmkp6eTtu2bStVV8OdIiJSI3JycmjcuLESNDmqmRmNGzc+pB5jJWkiIlJjlKBJbXCon3MlaSIiUiulpaWRmJhY6rFrrrmGVatWVfqckydPZvz48QC88MILvP766+WWX7JkCTfddFOlrwMwaNAglixZUm4Mh+qJJ55g3759ZR7PyMggMjKS//znPwftj48/+P2ixWN5/fXXSUxMpGvXrnTp0oVHH320SnECzJo1i44dO3LCCSfw8MMPl1pm586dnH/++SQlJdGnTx9SU1N9x5588klfTE888YRv/5133klSUhLdu3dn2LBhbNq0CfDMpRwzZgzdunWjc+fOPPTQQ1W+h7IoSRMRESnm5ZdfpkuXLlU6x/XXX88VV1xRbplevXrx1FNPVek6h1tBQUGFSdo777zDySefTEpKSsDn/fjjj3niiSeYPXs2P/zwA9999x3169evcqw33ngjH3/8MatWrSIlJaXU5PrBBx+ke/furFixgtdff52bb74ZgNTUVF566SUWLVrE8uXL+fDDD1m7di0AEyZMYMWKFSxbtoyzzz6b++67z3fvBw4cYOXKlSxdupT//Oc/pKWlVek+yqIkTUREaq38/HzGjBlDUlISF154oS8x8e+lSklJoVu3biQmJjJx4kRf3VmzZtGzZ0+Sk5MZMmRIiXPfc889vp6iQYMGMXHiRPr06UOHDh348ssvAZg3bx5nn302AHv37uWqq66iW7duJCUl8e677wIwbtw4evXqRdeuXbn77rsDuq8NGzZwxhln0LFjR+69917f/jfeeIM+ffrQvXt3rrvuOgoKCgBPD9hdd93FSSedxAMPPMCmTZsYPHgwgwcPLvX8KSkpPPbYY6Snp7Nx48aAYnrooYd49NFHad68OeB54vFPf/pTQHXLsmjRIk444QTatWtHVFQUo0eP5v333y9RbtWqVb6fUadOnUhLS2Pr1q2sXr2ak08+mdjYWCIiIhg4cCDTp08HoF69er762dnZviFLMyM7O5v8/Hz2799PVFTUQWUPJz3dKSIiNc4dmA8F2w/vScObYHUGlFtkzZo1vPLKK/Tr14+xY8fy3HPPcdttt/mOb9q0iYkTJ7J06VIaNmzIsGHDmDFjBv369eNPf/oT8+fPp23btuzYsaPCcPLz81m0aBEzZ87k3nvv5dNPPz3o+D//+U/q16/PypUrAc8QHcADDzxAo0aNKCgoYMiQIaxYsYKkpKRyr7Vo0SJSU1OJjY2ld+/ejBgxgri4OKZOncqCBQuIjIzkhhtu4M033+SKK64gOzubxMREX2/Rq6++yty5c2nSpEmJc2/YsIEtW7bQp08fLrroIqZOncott9xS4f2npqZy4oknVljuzTff5JFHHimx/4QTTmDatGkH7du4cSMtW7b0bbdo0YJvv/22RN3k5GTee+89+vfvz6JFi1i3bh3p6ekkJiZyxx13kJmZSUxMDDNnzqRXr16+enfccQevv/469evXZ+7cuQBceOGFvP/++zRr1ox9+/bx+OOP06hRowrv61CoJ01ERGqtli1b0q9fPwAuv/xyvvrqq4OOL168mEGDBpGQkEBERASXXXYZ8+fPZ+HChQwYMMD3SoVAfklfcMEFAJx44omlDo99+umn3Hjjjb7thg0bAvD222/Ts2dPevTowQ8//BDQXLmhQ4fSuHFjYmJiuOCCC/jqq6/47LPPWLp0Kb1796Z79+589tln/PrrrwCEh4czatSoCs8LMGXKFC666CIARo8eXeGQZ2UnzV922WUsW7asxFfxBA08r7cI5Hq33347O3fupHv37jz99NP06NGDiIgIOnfuzMSJExk6dChnnHEGycnJRET83n/1wAMPsGHDBi677DKeeeYZwJMAh4eHs2nTJn777Tcee+wxXzsebupJExGRGldRj1fQrlvsF3rx7dKSgKL9lU0+6tSpA3gSovz8/IDO+dtvv/Hoo4+yePFiGjZsyJVXXlniVQ7Tp0/3DWm+/PLLpd6HmeGcY8yYMaVOdI+OjiY8PDyg+0hJSWHr1q28+eabgKe3ce3atbRv356YmBhyc3OJiooCYMeOHb7euK5du7J06VJOO+20cs9fmZ60Fi1asGHDBt92enq6bzjVX7169fjvf/8LeNq5bdu2vgT76quv5uqrrwbg73//Oy1atChR/9JLL2XEiBHce++9vPXWW5xxxhlERkbStGlT+vXrx5IlS2jXrl2593Uo1JMmIiK11vr16/nmm28AT/LRv3//g46fdNJJfPHFF2zfvp2CggJSUlIYOHAgffv25YsvvuC3334DCGi4syLDhg3z9daAZ7hzz549xMXFUb9+fbZu3crHH39cot7555/v620qGqqbM2cOO3bsYP/+/b7h2SFDhjBt2jS2bdvmi3ndunWlxlK3bl2ysrJK7F+zZg3Z2dls3LiRtLQ00tLSmDRpElOmTAFg4MCBvPHGGwDs37+ft99+2zevbdKkSfztb39jy5YtABw4cKDUhyYq05PWu3dv1q5dy2+//UZubi5Tpkzh3HPPLVFu165d5ObmAp5EdsCAAb55ZEXtsX79et577z0uueQSAN8DBAAffPABnTp1AqBVq1Z8/vnnOOfIzs5m4cKFvmOHm3rSRESk1urcuTOvvfYa1113He3bt2fcuHEHHW/WrBkPPfQQgwcPxjnHWWedxciRIwF48cUXueCCCygsLKRp06bMmTOnSrH84x//4MYbbyQxMZHw8HDuvvtuLrjgAnr06EHXrl1p166db2i2Iv379+ePf/wjP//8M5deeqkvebv//vsZNmwYhYWFREZG8uyzz9K6desS9a+99lrOPPNMmjVr5puLBZ5E9vzzzz+o7KhRoxg9ejR33nknTz75JNdddx1PPfUUzjmuuOIKBgzw9JKeddZZbN26ldNPP93Xazh27NhDbS4AIiIieOaZZxg+fDgFBQWMHTuWrl27Ap5XoIDnKdvVq1dzxRVXEB4eTpcuXXjllVcOij8zM9PXHkXDzLfffjtr1qwhLCyM1q1b+8534403ctVVV5GYmIhzjquuuqrCOYKHysrqyj0S9erVy5X2zhiAGU9/zLM3v8rIG8/g2HEnATCqYxNmfpICUT0YcfogALYf2M6XGQvon3AKCXUSqit0EZFaZ/Xq1XTu3LmmwxCpFqV93s1sqXOuVxlVNNwpIiIiEoqCmqSZ2RlmtsbMfjaz20s5PsHMlnm/Us2swMwaeY+lmdlK77HSu8dEREREjlJBm5NmZuHAs8BQIB1YbGYfOOd8zw475x4BHvGWPwf4q3POf/blYOfcYX5xjoiIiEjoC2ZPWh/gZ+fcr865XGAKMLKc8pcAga8vISIiInIUC2aSdhywwW873buvBDOLBc4A3vXb7YDZZrbUzK4NWpQiIiIiISiYr+Ao7S1/ZT1Keg6woNhQZz/n3CYzawrMMbMfnXPzS1zEk8BdC553l4iIiIgcDYLZk5YOtPTbbgFsKqPsaIoNdTrnNnn/3AZMxzN8WoJz7kXnXC/nXK+EBL0yQ0REApOWlkZiYmKpx6655poKl1+Kj4+v0vXnzZvH119/XW6ZkSNH0rdv34P2XXnllSVe7Oofy08//cRZZ53FCSecQOfOnbnooovYunVrlWLdsWMHQ4cOpX379gwdOtS3rmhxTz75JImJiXTt2pUnnnjCt3/58uX07duXbt26cc4557Bnz56D6q1fv574+HjfgvSAb5mmrl27cv311/sWg69NgpmkLQbam1lbM4vCk4h9ULyQmdUHBgLv++2LM7O6Rd8Dw4DUIMYqIiLi8/LLL9OlS5egnT8/P7/CJG3Xrl1899137Nq1y7eyQUVycnIYMWIE48aN4+eff2b16tWMGzeOjIyMKsX78MMPM2TIENauXcuQIUN4+OGHS5RJTU3lpZdeYtGiRSxfvpwPP/zQ99b+a665hocffpiVK1dy/vnnl1j26a9//StnnnnmQfvefvttli9fTmpqKhkZGbzzzjtVuocjUdCSNOdcPjAe+ARYDbztnPvBzK43s+v9ip4PzHbOZfvtOwb4ysyWA4uAj5xzs4IVq4iI1E75+fmMGTOGpKQkLrzwQvbt2wfAoEGDKHo5ekpKCt26dSMxMZGJEyceVP/WW2+lZ8+eDBkyxJcI/fLLL5xxxhmceOKJnHrqqfz444+ApwfslltuYfDgwVx88cW88MILPP7443Tv3p0vv/yyRGzvvvsu55xzDqNHj/Ytu1SRt956i759+3LOOef49g0ePLjMHsNAvf/++4wZMwaAMWPGMGPGjBJlVq9ezcknn0xsbCwREREMHDiQ6dOnA57lpIpWHhg6dCjvvvv7FPQZM2bQrl0730oBRYqWbcrPzyc3N7fSa6UeDYK6LJRzbiYws9i+F4ptTwYmF9v3K5AczNhERCR0rNi1kt15uw/rOetH1iepQbdyy6xZs4ZXXnmFfv36MXbsWJ577jluu+023/FNmzYxceJEli5dSsOGDRk2bBgzZszgvPPOIzs7m549e/LYY49x3333ce+99/LMM89w7bXX8sILL9C+fXu+/fZbbrjhBj7//HPAMxT56aefEh4ezj333EN8fPxB1/OXkpLC3XffzTHHHMOFF17IpEmTKrzn1NRUTjzxxArLZWVlceqpp5Z67K233irRi7h161aaNWsGeJbKKlrv0l9iYiJ33HEHmZmZxMTEMHPmTN9yVImJiXzwwQeMHDmSd955x7coenZ2Nv/617+YM2fOQUOdRYYPH86iRYs488wzufDCCyu8r6ONVhwQEZFaq2XLlr71MC+//HK++uqrg44vXryYQYMGkZCQQEREBJdddhnz53ueYQsLC+Piiy8+qO7evXv5+uuv+cMf/kD37t257rrr2Lx5s+98f/jDHwgPD68wrq1bt/Lzzz/Tv39/OnToQEREBKmpnlk/pfUoVbaXqW7duqUuYr5s2bJDHubt3LkzEydOZOjQob75ZBERnr6gV199lWeffZYTTzyRrKwsoqKiALj77rv561//Wub8vk8++YTNmzdz4MABX6Jbm2iBdRERqXEV9XgFS/Hkpvh2Zda3NjMKCwtp0KABy5YtK7VMXFxcQOeaOnUqO3fupG3btgDs2bOHKVOmcP/999O4ceODJu7v2LGDJk2aANC1a1e++OKLCs9f2Z60Y445hs2bN9OsWTM2b95M06ZNS6179dVXc/XVVwPw97//nRYtWgDQqVMnZs+eDXh6Ez/66CMAvv32W6ZNm8bf/vY3du3aRVhYGNHR0YwfP953zujoaM4991zef/99hg4dWuG9HU3UkyYiIrXW+vXr+eabbwDP8GL//v0POn7SSSfxxRdfsH37dgoKCkhJSWHgwIEAFBYW+p6yfOutt+jfvz/16tWjbdu2vknuzjmWL19e6rXr1q1LVlZWqcdSUlKYNWsWaWlppKWlsXTpUt+8tEGDBjF16lRyc3MBmDx5MoMHDwbg0ksv5euvv/YlQQCzZs1i5cqVJa5dmZ60c889l9deew2A1157jZEjS383fdEw6Pr163nvvfe45JJLDtpfWFjI/fffz/XXe6amf/nll757/Mtf/sLf//53xo8fz969e309kPn5+cycOZNOnTqVes2jmZI0ERGptTp37sxrr71GUlISO3bsYNy4cQcdb9asGQ899BCDBw8mOTmZnj17+hKUuLg4fvjhB0488UQ+//xz7rrrLgDefPNNXnnlFd/rI95///0S1wU455xzmD59eokHB9LS0li/fj0nn3yyb1/btm2pV68e3377LWeffTannnoqJ554It27d2fBggX861//AiAmJoYPP/yQp59+mvbt29OlSxcmT55cZs9XoG6//XbmzJlD+/btmTNnDrff7lmOe9OmTZx11lm+cqNGjaJLly6cc845PPvsszRs2BDwJJ0dOnSgU6dONG/enKuuuqrc62VnZ3PuueeSlJREcnIyTZs29SV2tYlVpis31PXq1csVPY1T3IynP+bZm19l5I1ncOy4kwAY1bEJMz9JgagejDh9EADbD2zny4wF9E84hYQ6eu+aiEiwrF69ms6dO9d0GCLVorTPu5ktdc71KquOetJEREREQpCSNBEREZEQpCRNREREJAQpSRMREREJQUrSREREREKQkjQRERGREKQkTUREpAJLlizhpptuAuCee+4pdZ3JynjwwQfLPf79999jZnzyySe+fWlpaSUWSi8ey6OPPkqnTp1ITEwkOTmZ119/vUpxguflte3bt6d9+/a+F9oWt27dOoYMGUJSUhKDBg0iPT3dd2zixIkkJiaSmJjI1KlTffudc9xxxx106NCBzp0789RTT/n233TTTZxwwgkkJSXx3XffVfkejlRK0kRERCrQq1cvXxJRFc45CgsLK0zSilY/SElJCfjcL7zwAnPmzGHRokWkpqYyf/78Si1rVZodO3Zw77338u2337Jo0SLuvffeg5akKnLbbbdxxRVXsGLFCu666y7fYvAfffQR3333HcuWLePbb7/lkUceYc+ePYBnpYQNGzbw448/snr1akaPHg3Axx9/zNq1a1m7di0vvvhiiRcM1yZK0kREpFbKzs5mxIgRJCcnH9TLs3jxYk455RSSk5Pp06cPWVlZzJs3j7PPPttXd/ny5Zx22mm0b9+el156ybf/kUceoXfv3iQlJXH33XcDnh6wzp07c8MNN9CzZ0+uvvpq9u/fT/fu3bnssstKxOWcY9q0aUyePJnZs2eTk5MT0P08+OCDPPfcc9SrVw+A+vXrM2bMmENuH/AscD506FAaNWpEw4YNGTp0KLNmzSpRbtWqVQwZMgSAwYMH+1ZZWLVqFQMHDiQiIoK4uDiSk5N99Z9//nnuuusuwsI8qUjRqgjvv/8+V1xxBWbGySefzK5duw5apL420QLrIiJS475I+42MfdmH9ZwJsXEMbNO2zOOzZs2iefPmvnUud+/eTW5uLhdffDFTp06ld+/e7Nmzh5iYmBJ1V6xYwcKFC8nOzqZHjx6MGDGC1NRU1q5dy6JFi3DOce655zJ//nxatWrFmjVr+O9//8tzzz0HwDvvvFPmIuwLFiygbdu2HH/88QwaNIiZM2dywQUXlHuvWVlZZGVlcfzxx1fYLo888ghvvvlmif0DBgwo0Vu4ceNGWrZs6dtu0aIFGzduLFE3OTmZd999l5tvvpnp06eTlZVFZmYmycnJ3Hvvvdxyyy3s27ePuXPn+tYG/eWXX5g6dSrTp08nISGBp556ivbt25d5zWbNmlV4b0cbJWkiIlIrdevWjdtuu42JEyf61sNcuXIlzZo1o3fv3gC+XqniRo4cSUxMDDExMQwePJhFixbx1VdfMXv2bHr06AHA3r17Wbt2La1ataJ169YHrcVZnpSUFN/Q3+jRo/nf//7HBRdcgJmVWt7McM6Veby4CRMmMGHChIDKljZcWtp1Hn30UcaPH8/kyZMZMGAAxx13HBEREQwbNszXM5mQkEDfvn2JiPCkHgcOHCA6OpolS5bw3nvvMXbsWL788suAr1kbKEkTEZEaV16PV7B06NCBpUuXMnPmTCZNmsSwYcM477zzAkoIipcpSpQmTZrEddddd9CxtLQ04uLiAoqpoKCAd999lw8++IAHHngA5xyZmZlkZWXRuHHjEvPBduzY4Vt8PS4ujl9//ZV27dqVe43K9KS1aNGCefPm+bbT09MZNGhQibrNmzfnvffeAzzJ6bvvvkv9+vUBuOOOO7jjjjsAuPTSS2nfvr3v3KNGjQLg/PPP9y263qJFCzZs2HDQNZs3b17uPR2tNCdNRERqpU2bNhEbG8vll1/ObbfdxnfffUenTp3YtGkTixcvBjzDiPn5+SXqvv/+++Tk5JCZmcm8efPo3bs3w4cP59VXX2Xv3r2AZ6hw27ZtpV47MjKSvLy8Evs//fRTkpOT2bBhA2lpaaxbt45Ro0YxY8YM4uPjadasGZ999hngSdBmzZpF//79AZg0aRI33nijb2L+nj17ePHFF0tcY8KECSxbtqzEV2kPRgwfPpzZs2ezc+dOdu7cyezZsxk+fHiJctu3b6ewsBCAhx56iLFjxwKepDMzMxPwDBGvWLGCYcOGAXDeeefx+eefA/DFF1/QoUMHAM4991xef/11nHMsXLiQ+vXr18qhTlBPmoiI1FIrV65kwoQJhIWFERkZyfPPP09UVBRTp07lz3/+M/v37ycmJoZPP/20RN0+ffowYsQI1q9fz5133knz5s1p3rw5q1evpm/fvgDEx8fzxhtvEB4eXqL+tddeS1JSEj179jyoVyslJYXzzz//oLKjRo3i+eef549//COvv/46N954I7feeisAd999t28e2rhx49i7dy+9e/cmMjKSyMhIX7lD1ahRI+68807f8O9dd91Fo0aNfN/36tWLc889l3nz5jFp0iTMjAEDBvDss88CkJeXx6mnngp4ho7feOMN33Dn7bffzmWXXcbjjz9OfHw8L7/8MgBnnXUWM2fO5IQTTiA2Npb//ve/VbqHI5lV9fHcUNKrVy+3ZMmSUo/NePpjnr35VUbeeAbHjjsJgFEdmzDzkxSI6sGI0wcBsP3Adr7MWED/hFNIqJNQXaGLiNQ6q1evpnPnzjUdhki1KO3zbmZLnXO9yqqj4U4RERGREKQkTURERCQEKUkTERERCUFK0kRERERCkJI0ERERkRCkJE1EREQkBClJExERCdA999zDo48+GvTrtGnThu3btwfl+g8++GC5x7///nvMjE8++cS3Ly0tjcTExHJjefTRR+nUqROJiYkkJyfz+uuvVylOgNdee4327dvTvn17XnvttVLLrFu3jiFDhpCUlMSgQYNIT0/3HZs4cSKJiYkkJiYydepU3/5TTz2V7t270717d5o3b855550HwLx586hfv77v2H333Vfle6gKvcxWRESkFnDO4ZzjwQcf5O9//3uZ5VJSUujfvz8pKSmlri5QmhdeeIE5c+awaNEi6tWrx+7du5kxY0aV4t2xYwf33nsvS5Yswcw48cQTOffcc2nYsOFB5W677TauuOIKxowZw+eff86kSZP43//+x0cffcR3333HsmXLOHDgAAMHDuTMM8+kXr16fPnll776o0aNYuTIkb7tU089lQ8//LBKsR8u6kkTEZFaKS0tjU6dOjFmzBiSkpK48MIL2bdvH3BwT9aSJUsOWq9y+fLlnHbaabRv356XXnoJ8PTADBw4kIsuuogOHTpw++238+abb9KnTx+6devGL7/8AkBGRgajRo2id+/e9O7dmwULFgCQmZnJsGHD6NGjB9ddd12pi4yXd33wrMnZu3dvkpKSuPvuu3332LlzZ2644QZ69uzJ1Vdfzf79++nevTuXXXZZiXM755g2bRqTJ09m9uzZ5OTkBNSWDz74IM8995xvQfr69eszZsyYgOqW5ZNPPmHo0KE0atSIhg0bMnToUGbNmlWi3KpVqxgyZAgAgwcP5v333/ftHzhwIBEREcTFxZGcnFyiflZWFp9//rmvJy3UqCdNRERq3MLvf2PHzuzDes5GDeM4uUf5C7evWbOGV155hX79+jF27Fiee+45brvttnLrrFixgoULF5KdnU2PHj0YMWIE4EmeVq9eTaNGjWjXrh3XXHMNixYt4sknn+Tpp5/miSee4Oabb+avf/0r/fv3Z/369QwfPpzVq1dz77330r9/f+666y4++uijUtfcLO/6qamprF27lkWLFuGc49xzz2X+/Pm0atWKNWvW8N///pfnnnsOgHfeeYdly5aVeu4FCxbQtm1bjj/+eAYNGsTMmTO54IILym2PrKwssrKyfMtTlacyi7tv3LiRli1b+rZbtGjBxo0bS9RNTk7m3Xff5eabb2b69OlkZWWRmZlJcnIy9957L7fccgv79u1j7ty5dOnS5aC606dPZ8iQIb7kEuCbb74hOTmZ5s2b8+ijj9K1a9cK7ytYlKSJiEit1bJlS/r16wfA5ZdfzlNPPVVhkjZy5EhiYmKIiYlh8ODBLFq0iAYNGtC7d2/fQuDHH3+8byHxbt26MXfuXMCzgPqqVat859qzZw9ZWVnMnz+f9957D4ARI0aUGNKr6PpfffUVs2fPpkePHgDs3buXtWvX0qpVK1q3bs3JJ58cUHukpKQwevRoAEaPHs3//vc/LrjgAsys1PJmhnOuzOPFTZgwgQkTJgRUtrTexNKu8+ijjzJ+/HgmT57MgAEDOO6444iIiGDYsGEsXryYU045hYSEBPr27etbN7RISkoK11xzjW+7Z8+erFu3jvj4eGbOnMl5553H2rVrA4o3GJSkiYhIjauoxytYiv/SL9qOiIigsLAQoMSQX1l16tSp49sXFhbm2w4LCyM/Px+AwsJCvvnmG2JiYiqMBeDZZ5/1DWnOnDmzzOs755g0aRLXXXfdQcfS0tKIi4srcd7SFBQU8O677/LBBx/wwAMP4JwjMzOTrKwsGjduzM6dOw8qv2PHDtq2bUu9evWIi4vj119/pV27duVeozI9aS1atGDevHm+7fT09IOGnYs0b97cl+Du3buXd999l/r16wNwxx13cMcddwBw6aWX0r59e1+9zMxMFi1axPTp0337/HvUzjrrLG644Qa2b99OkyZNyr2vYNGcNBERqbXWr1/PN998A/w+YR48c9KWLl0KwLvvvntQnffff5+cnBwyMzOZN28evXv3Dvh6w4YN45lnnvFtFw07DhgwwJe8fPzxx76E6MYbb2TZsmUsW7aM5s2bl3n94cOH8+qrr7J3717AM1S4bdu2UmOIjIwkLy+vxP5PP/2U5ORkNmzYQFpaGuvWrWPUqFHMmDGD+Ph4mjVrxmeffQZ4ErRZs2b52mvSpEnceOON7NmzB/D0EJY2ZDthwgTf/fh/FU/QAIYPH87s2bPZuXMnO3fuZPbs2aU+yLB9+3ZfQv3QQw8xduxYwJN0ZmZmAp4h4hUrVvh6N8Ez7Hv22WcTHR3t27dlyxZfD96iRYsoLCykcePGpbZjdVCSJiIitVbnzp157bXXSEpKYseOHYwbNw6Au+++m5tvvplTTz2V8PDwg+r06dOHESNGcPLJJ3PnnXf6kqdAPPXUUyxZsoSkpCS6dOnCCy+84Lve/Pnz6dmzJ7Nnz6ZVq1ZlnqO06w8bNoxLL72Uvn370q1bNy688EKysrJKrX/ttdeSlJRU4sGBlJQUzj///IP2jRo1irfeeguA119/nfvvv5/u3btz2mmncffdd/vmoY0bN47BgwfTu3dvEhMTGThwILGxsQG3S2kaNWrEnXfe6XvI4q677qJRo0YA3HXXXXzwwQeA56GNjh070qFDB7Zu3errOcvLy+PUU0+lS5cuXHvttbzxxhsHDXdOmTKFSy655KBrTps2zfcKkZtuuokpU6YEPJQbDFbeEyRHml69erklS5aUemzG0x/z7M2vMvLGMzh23EkAjOrYhJmfpEBUD0acPgiA7Qe282XGAvonnEJCnYTqCl1EpNZZvXo1nTt3rrHrp6WlcfbZZ5OamlpjMUjtUdrn3cyWOud6lVVHPWkiIiIiIUhJmoiI1Ept2rRRL5qENCVpIiIiIiFISZqIiIhICFKSJiIiIhKClKSJiIiIhCAlaSIiUiulpaWRmJhY6rFrrrnmoOWbgu3KK69k2rRph1x/8uTJjB8/vkoxPPHEE74F5isjPz+fJk2aMGnSpIP2+y9SD573mZ199tm+7Y8//phevXrRuXNnOnXqVOFyXIFYunQp3bp144QTTuCmm24qdWmp3NxcrrrqKrp160ZycvJBqxpMnTqVpKQkunbtyt/+9jff/vXr1zN48GB69OhBUlKSb/WHuXPn0r17d99XdHQ0M2bMqPJ9FFGSJiIiUszLL79cYjHumlZQUBDUcx9qkjZ79mw6duzI22+/XWpSVJrU1FTGjx/PG2+8werVq0lNTa1wSalAjBs3jhdffJG1a9eydu1aZs2aVaJM0TJbK1euZM6cOdx6660UFhaSmZnJhAkT+Oyzz/jhhx/YunWrb4WF+++/n4suuojvv/+eKVOmcMMNNwAwePBg36oJn3/+ObGxsQetalBVStJERKTWys/PZ8yYMSQlJXHhhRf6kpRBgwZR9HL0lJQUunXrRmJiIhMnTvTVjY+PZ+LEiZx44omcfvrpLFq0iEGDBtGuXTvf2/ALCgqYMGECvXv3Jikpif/85z+AZ/Hw8ePH06VLF0aMGFHmEk7z5s1j8ODBXHrppXTr1o2cnBxfL1CPHj18C7cDbNiwgTPOOIOOHTty7733+va/8cYb9OnTh+7du3Pdddf5kr34+HjuuusuTjrpJB544AE2bdrE4MGDGTx4MOBJeHr16kXXrl25++67y2zDlJQUbr75Zlq1asXChQsDavf/9//+H3fccQedOnUCPGulFiU+h2rz5s3s2bOHvn37YmZcccUVpfZqrVq1iiFDhgDQtGlTGjRowJIlS/j111/p0KEDCQmeF9mffvrpviXBzMy35NXu3btLXWVi2rRpnHnmmVVeacGfFlgXEZEaNzTsD0E575zCd8o9vmbNGl555RX69evH2LFjee655w4adtu0aRMTJ05k6dKlNGzYkGHDhjFjxgzOO+88srOzGTRoEP/61784//zz+cc//sGcOXNYtWoVY8aM4dxzz+WVV16hfv36LF68mAMHDtCvXz+GDRvG999/z5o1a1i5ciVbt26lS5cuvjUni1u0aBGpqam0bduWxx57DPD0Av34448MGzaMn3766aBysbGx9O7dmxEjRhAXF8fUqVNZsGABkZGR3HDDDbz55ptcccUVZGdnk5iYyH333QfAq6++yty5c32LiT/wwAM0atSIgoIChgwZwooVK0hKSjootv379/PZZ5/xn//8h127dpGSkkLfvn0r/LmkpqZy6623Vlhu7ty5/PWvfy2xPzY2lq+//vqgfRs3bqRFixa+7RYtWrBx48YSdZOTk3n//fcZPXo0GzZsYOnSpWzYsIHTTjuNH3/8kbS0NFq0aMGMGTPIzc0F4J577mHYsGE8/fTTZGdn8+mnn5Y475QpU7jlllsqvKfKUJImIiK1VsuWLenXrx8Al19+OU899dRBSdrixYsZNGiQr3flsssuY/78+Zx33nlERUVxxhlnANCtWzfq1KlDZGQk3bp1Iy0tDfAMBa5YscI332z37t2sXbuW+fPnc8kllxAeHk7z5s057bTTyoyxT58+tG3bFoCvvvqKP//5zwB06tSJ1q1b+5K0oUOH+hYDv+CCC/jqq6+IiIhg6dKlvkXg9+/fT9OmTQEIDw9n1KhRZV737bff5sUXXyQ/P5/NmzezatWqEknahx9+yODBg4mNjWXUqFH885//5PHHHyc8PLzUNS8ruw5m0XBiIEobai3temPHjmX16tX06tWL1q1bc8oppxAREUHDhg15/vnnufjiiwkLC+OUU07h119/BTy9hVdeeSW33nor33zzDX/84x9JTU0lLMwzILl582ZWrlxZ6gLwVaEkTUREalxFPV7BUvyXePHt8uZYRUZG+sqHhYVRp04d3/f5+fm++k8//XSJX94zZ84sNYH49ttvue666wC47777qFevHnFxcQHFU9q9OOcYM2YMDz30UIny0dHRJRaPL/Lbb7/x6KOPsnjxYho2bMiVV15JTk5OiXIpKSksWLCANm3aAJCZmcncuXM5/fTTady4MTt37vT1zO3YscP3fdeuXVm6dCnJycll3g9UrietRYsWpKen+7bT09NLHZaMiIjg8ccf922fcsoptG/fHoBzzjmHc845B4AXX3zR1z6vvPKKb35b3759ycnJYfv27b6E9+233+b8888nMjKy3PupLM1JExGRWmv9+vV88803gCfh6N+//0HHTzrpJL744gu2b99OQUEBKSkpDBw4MODzDx8+nOeff568vDwAfvrpJ7KzsxkwYABTpkyhoKCAzZs3++aWnXTSSb6J6Oeee26J8w0YMIA333zTd67169fTsWNHAObMmcOOHTvYv38/M2bMoF+/fgwZMoRp06b55rzt2LGDdevWlRpr3bp1ycrKAmDPnj3ExcVRv359tm7dyscff1yi/J49e/jqq69Yv349aWlppKWl8eyzz5KSkgJ45vX973//Azxz89544w3ffLcJEybw4IMP+noBCwsL+fe//13iGv4T8/2/iidoAM2aNaNu3bosXLgQ5xyvv/46I0eOLFFu3759ZGdn+9osIiLC95BIUTvt3LmT5557jmuuuQaAVq1a+R4iWL16NTk5Ob7eVfB8di655JJS27Uq1JMmIiK1VufOnXnttde47rrraN++PePGjTvoeLNmzXjooYcYPHgwzjnOOuusUn/xl+Waa64hLS2Nnj174pwjISGBGTNmcP755/P555/TrVs3OnToEHDid8MNN3D99dfTrVs3IiIimDx5sq8Hr3///vzxj3/k559/5tJLL6VXr16A58nEYcOGUVhYSGRkJM8++yytW7cuce5rr72WM888k2bNmjF37lx69OhB165dadeunW9I2N97773Haaed5rs+wMiRI/nb3/7GgQMHuPPOOxk3bhzJyck45zjjjDO4/PLLAUhKSuKJJ57gkksuYd++fZgZI0aMCLhdy/L8889z5ZVXsn//fs4880zOPPNMAD744AOWLFnCfffdx7Zt2xg+fDhhYWEcd9xxvkQS4Oabb2b58uUA3HXXXXTo0AGAxx57jD/96U88/vjjmBmTJ0/29VympaWxYcOGSiXvgbJAH5c9EvTq1csVPY1T3IynP+bZm19l5I1ncOy4kwAY1bEJMz9JgagejDh9EADbD2zny4wF9E84hYQ6CaWeS0REqm716tV07ty5psMQqRalfd7NbKlzrldZdTTcKSIiIhKClKSJiIiIhCAlacUUuMKaDkFERERESZo/5xxr9/5MZFgk9SPr13Q4IiJHvaNpXrRIWQ71c64kzc/G/ZvIyMmgS73ORIVF1XQ4IiJHtejoaDIzM5WoyVHNOUdmZibR0dGVrqtXcHjlF+aTujuV+lH1aRNX8tFkERE5vIpePpqRkVHToYgEVXR09EFLVgVKSZrX2r0/s78gh96NehFm6mAUEQm2yMhI33JHIlKSshGvHbk7aBDVgMZ1Gtd0KCIiIiJK0vypB01ERERChbISERERkRB0VC0LZWYZQOkrx/6uCbC9GsKp7dTO1UPtXD3UztVD7Vw91M7VI5B2bu2cK3MNyqMqSQuEmS0pb50sOTzUztVD7Vw91M7VQ+1cPdTO1eNwtLOGO0VERERCkJI0ERERkRBUG5O0F2s6gFpC7Vw91M7VQ+1cPdTO1UPtXD2q3M61bk6aiIiIyJGgNvakiYiIiIS8oyZJM7MzzGyNmf1sZreXctzM7Cnv8RVm1jPQuvK7KrZzmpmtNLNlZrakeiM/sgTQzp3M7BszO2Bmt1Wmrvyuiu2sz3OAAmjny7z/Xqwws6/NLDnQuvK7KrazPs8BCqCdR3rbeJmZLTGz/oHWLcE5d8R/AeHAL0A7IApYDnQpVuYs4GPAgJOBbwOtq6+qt7P3WBrQpKbvI9S/AmznpkBv4AHgtsrU1VfV29l7TJ/nw9fOpwANvd+fqX+fq7edvdv6PB++do7n9+lkScCPgdYt/nW09KT1AX52zv3qnMsFpgAji5UZCbzuPBYCDcysWYB1xaMq7SyBq7CdnXPbnHOLgbzK1hWfqrSzBC6Qdv7aObfTu7kQaBFoXfGpSjtL4AJp573Om5UBcYALtG5xR0uSdhywwW873bsvkDKB1BWPqrQzeD6os81sqZldG7Qoj3xV+Uzq8xy4qraVPs+BqWw7X42nN/5Q6tZmVWln0Oc5UAG1s5mdb2Y/Ah8BYytT119ElUINHVbKvuKPrZZVJpC64lGVdgbo55zbZGZNgTlm9qNzbv5hjfDoUJXPpD7PgatqW+nzHJiA29nMBuNJHorm8OjzHLiqtDPo8xyogNrZOTcdmG5mA4B/AqcHWtff0dKTlg609NtuAWwKsEwgdcWjKu2Mc67oz23AdDxdv1JSVT6T+jwHrkptpc9zwAJqZzNLAl4GRjrnMitTV4CqtbM+z4Gr1GfSm+geb2ZNKlsXjp4kbTHQ3szamlkUMBr4oFiZD4ArvE8fngzsds5tDrCueBxyO5tZnJnVBTCzOGAYkFqdwR9BqvKZ1Oc5cIfcVvo8V0qF7WxmrYD3gD86536qTF3xOeR21ue5UgJp5xPMzLzf98TzkEBmIHWLOyqGO51z+WY2HvgEz9MTrzrnfjCz673HXwBm4nny8GdgH3BVeXVr4DZCXlXaGTgGT9cveD53bznnZlXzLRwRAmlnMzsWWALUAwrN7C94nhLao89zYKrSzkAT9HkOSID/btwFNAae87ZpvnOul/59DlxV2hn9+xywANt5FJ7OijxgP3Cx90GCSn+eteKAiIiISAg6WoY7RURERI4qStJEREREQpCSNBEREZEQpCRNREREJAQpSRMREREJQUrSRKTamFmBmS3z+2pTTtm9h+F6k83sN++1vjOzvodwjpfNrIv3+78XO/Z1VWP0nqeoXVLN7P/MrEEF5bub2VmH49oiErr0Cg4RqTZmttc5F3+4y5ZzjsnAh865aWY2DHjUOZdUhfNVOaaKzmtmrwE/OeceKKf8lUAv59z4wx2LiIQO9aSJSI0xs3gz+8zby7XSzEaWUqaZmc3362k61bt/mJl94637jplVlDzNB07w1r3Fe65U7wtqi966/pGZLffuv9i7f56Z9TKzh4EYbxxveo/t9f451b9ny9uDN8rMws3sETNbbGYrzOy6AJrlG7yLLptZHzP72sy+9/7Z0fum8vuAi72xXOyN/VXvdb4vrR1F5MhzVKw4ICJHjBgzW+b9/jfgD8D53pUSmgALzewDd3AX/6XAJ865B8wsHIj1lv0HcLpzLtvMJgK34EleynIOsNLMTsSzEsZJeBY8/tbMvgDaAZuccyMAzKy+f2Xn3O1mNt45172Uc08BLgZmepOoIcA4PItY73bO9TazOsACM5vtnPuttAC99zcEeMW760dggPct56cDDzrnRpnZXfj1pJnZg8Dnzrmx3qHSRWb2qXMuu5z2EJEQpyRNRKrTfv8kx8wigQfNbABQiKcH6Rhgi1+dxcCr3rIznHPLzGwgnuWZFniXsonC0wNVmkfM7B9ABp6kaQgwvSiBMbP3gFOBWcCjZvYvPEOkX1bivj4GnvImYmcA851z+71DrElmdqG3XH2gPZ4E1V9R8toGWArM8Sv/mpm1BxwQWcb1hwHnmtlt3u1ooBWwuhL3ICIhRkmaiNSky4AE4ETnXJ6ZpeFJMHycc/O9SdwI4H9m9giwE5jjnLskgGtMcM5NK9rw9kiV4Jz7ydvLdhbwkLfHq7yeOf+6OWY2DxiOp0ctpehywJ+dc59UcIr9zrnu3t67D4EbgaeAfwJznXPnex+ymFdGfQNGOefWBBKviBwZNCdNRGpSfWCbN0EbDLQuXsDMWnvLvIRnGLAnsBDoZ2ZFc8xizaxDgNecD5znrRMHnA98aWbNgX3OuTeAR73XKS7P26NXmil4hlFPxbOAMt4/xxXVMbMO3muWyjm3G7gJuM1bpz6w0Xv4Sr+iWUBdv+1PgD+bt1vRzHqUdQ0ROXIoSRORmvQm0MvMluDpVfuxlDKDgGVm9j0wCnjSOZeBJ2lJMbMVeJK2ToFc0Dn3HTAZWAR8C7zsnPse6IZnLtcy4A7g/lKqvwisKHpwoJjZwADgU+dcrnffy8Aq4DszSwX+QwUjGN5YlgOjgf+Hp1dvARDuV2wu0KXowQE8PW6R3thSvdsicoTTKzhEREREQpB60kRERERCkJI0ERERkRCkJE1EREQkBClJExEREQlBStJEREREQpCSNBEREZEQpCRNREREJAQpSRMREREJQf8fVHB8v7rK0MMAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_roc(probs_df, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "b29c522b",
   "metadata": {},
   "outputs": [],
   "source": [
    "val = pd.read_csv('./data/Validation.tsv', sep='\\t')\n",
    "val['concat'] = val.Title.map(str) + \" \" + val.Abstract.fillna(' ').map(str)\n",
    "val['bert'] = val['concat'].apply(lambda x: x.lower())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "e04a0823",
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer = AutoTokenizer.from_pretrained('allenai/biomed_roberta_base', do_lower_case=True)\n",
    "\n",
    "val_inputs, val_masks = preprocessing_for_bert(tokenizer, val.bert)\n",
    "\n",
    "val_data = TensorDataset(val_inputs, val_masks)\n",
    "val_sampler = SequentialSampler(val_data)\n",
    "val_dataloader = DataLoader(val_data, sampler=val_sampler, batch_size=batch_size, num_workers=1)\n",
    "\n",
    "bert_classifier = torch.load('./models/' + 'full-allenai-biomed_roberta_base.pkl')\n",
    "\n",
    "probs = bert_predict(bert_classifier, val_dataloader, device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "bc29175e",
   "metadata": {},
   "outputs": [],
   "source": [
    "preds = np.where(probs[:, 1] > 0.5, 1, 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "9f9c9b73",
   "metadata": {},
   "outputs": [],
   "source": [
    "pd.Series(preds).to_csv('submission-1.csv', index=False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}